SHBGF Algorithm Documentation

🔍 Overview

The Soft HBGF (sHBGF) is a consensus clustering algorithm that combines multiple soft clusterings by representing each data point with a concatenated membership vector and clustering them using KMeans. It is the soft version of the HBGF/hypergraph bipartite graph formulation.

⚙️ Class Definition

Class Name: SHBGF

class SHBGF:
    def __init__(self, n_clusters: int, max_iter: int = 10):
        self.n_clusters = n_clusters
        self.max_iter = max_iter

Implements the soft HBGF algorithm that merges membership vectors from multiple soft clusterings and applies KMeans to produce consensus labels.

📋 Parameters

Parameter	Type	Default	Description
`n_clusters`	int	—	Number of consensus clusters to output
`max_iter`	int	10	Maximum number of iterations for KMeans

💻 Using Example

from soft_clustering._shbgf._shbgf import SHBGF
import numpy as np

# Simulate 3 soft clusterings
soft1 = np.random.dirichlet(np.ones(3), size=100)
soft2 = np.random.dirichlet(np.ones(3), size=100)
soft3 = np.random.dirichlet(np.ones(3), size=100)

model = SHBGF(n_clusters=3)
labels = model.fit_predict([soft1, soft2, soft3])
print("Consensus Labels:", labels)

📥 Input / 📤 Output

Input to fit_predict(soft_memberships):
- soft_memberships (list of np.ndarray): List of soft clustering matrices (shape N x K)
Returns:
- labels (np.ndarray): Consensus clustering labels (length N)

🔧 Methods

__init__(self, n_clusters: int, max_iter: int = 10): Initializes the model with number of clusters and KMeans iterations.
fit_predict(self, soft_memberships: List[np.ndarray]): Concatenates membership vectors and applies KMeans to find consensus labels.

📝 Implementation Notes

This method treats each data point as a concatenation of soft cluster memberships across different sources.
The result is a single vector per object, which is clustered using KMeans.
It avoids transforming soft memberships into hard labels, preserving more probabilistic information.

📚 Reference

Punera, K., & Ghosh, J. (2008). Consensus-Based Ensembles of Soft Clusterings.